public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Andrew MacLeod <amacleod@redhat.com>
To: gcc-patches <gcc-patches@gcc.gnu.org>
Cc: "hernandez, aldy" <aldyh@redhat.com>
Subject: [PATCH 12/17] Convert CFN_CTZ builtins to range-ops.
Date: Thu, 22 Sep 2022 15:05:36 -0400	[thread overview]
Message-ID: <622e4a48-eae4-600f-db3c-c478f537caa7@redhat.com> (raw)
In-Reply-To: <571782f9-72e6-5c30-da55-b8d62d3a153e@redhat.com>

[-- Attachment #1: Type: text/plain, Size: 75 bytes --]

Bootstrapped on x86_64-pc-linux-gnu with no regressions.  Pushed.

Andrew

[-- Attachment #2: 0012-Convert-CFN_CTZ-builtins-to-range-ops.patch --]
[-- Type: text/x-patch, Size: 5945 bytes --]

From 55738d8d96bb4f39a72cf5e3739d35b39fc2146a Mon Sep 17 00:00:00 2001
From: Andrew MacLeod <amacleod@redhat.com>
Date: Tue, 20 Sep 2022 18:19:30 -0400
Subject: [PATCH 12/17] Convert CFN_CTZ builtins to range-ops.

	* gimple-range-fold.cc (range_of_builtin_int_call): Remove case
	for CFN_CTZ.
	* gimple-range-op.cc (class cfn_ctz): New.
	(gimple_range_op_handler::maybe_builtin_call): Set arguments.
---
 gcc/gimple-range-fold.cc | 61 +------------------------------
 gcc/gimple-range-op.cc   | 79 ++++++++++++++++++++++++++++++++++++++++
 2 files changed, 80 insertions(+), 60 deletions(-)

diff --git a/gcc/gimple-range-fold.cc b/gcc/gimple-range-fold.cc
index 63eaa90be96..96a138a7a02 100644
--- a/gcc/gimple-range-fold.cc
+++ b/gcc/gimple-range-fold.cc
@@ -917,7 +917,7 @@ fold_using_range::range_of_builtin_int_call (irange &r, gcall *call,
 
   tree type = gimple_range_type (call);
   tree arg;
-  int mini, maxi, zerov = 0, prec;
+  int prec;
   scalar_int_mode mode;
 
   switch (func)
@@ -926,65 +926,6 @@ fold_using_range::range_of_builtin_int_call (irange &r, gcall *call,
       r.set (build_zero_cst (type), build_one_cst (type));
       return true;
 
-    CASE_CFN_CTZ:
-      // __builtin_ctz* return [0, prec-1], except for when the
-      // argument is 0, but that is undefined behavior.
-      //
-      // For __builtin_ctz* consider argument of 0 always undefined
-      // behavior, for internal fns depending on CTZ_DEFINED_VALUE_AT_ZERO.
-      arg = gimple_call_arg (call, 0);
-      prec = TYPE_PRECISION (TREE_TYPE (arg));
-      mini = 0;
-      maxi = prec - 1;
-      mode = SCALAR_INT_TYPE_MODE (TREE_TYPE (arg));
-      if (gimple_call_internal_p (call))
-	{
-	  if (optab_handler (ctz_optab, mode) != CODE_FOR_nothing
-	      && CTZ_DEFINED_VALUE_AT_ZERO (mode, zerov) == 2)
-	    {
-	      // Handle only the two common values.
-	      if (zerov == -1)
-		mini = -1;
-	      else if (zerov == prec)
-		maxi = prec;
-	      else
-		// Magic value to give up, unless we can prove arg is non-zero.
-		mini = -2;
-	    }
-	}
-      src.get_operand (r, arg);
-      if (!r.undefined_p ())
-	{
-	  // If arg is non-zero, then use [0, prec - 1].
-	  if (!range_includes_zero_p (&r))
-	    {
-	      mini = 0;
-	      maxi = prec - 1;
-	    }
-	  // If some high bits are known to be zero, we can decrease
-	  // the maximum.
-	  wide_int max = r.upper_bound ();
-	  if (max == 0)
-	    {
-	      // Argument is [0, 0].  If CTZ_DEFINED_VALUE_AT_ZERO
-	      // is 2 with value -1 or prec, return [-1, -1] or [prec, prec].
-	      // Otherwise ignore the range.
-	      if (mini == -1)
-		maxi = -1;
-	      else if (maxi == prec)
-		mini = prec;
-	    }
-	  // If value at zero is prec and 0 is in the range, we can't lower
-	  // the upper bound.  We could create two separate ranges though,
-	  // [0,floor_log2(max)][prec,prec] though.
-	  else if (maxi != prec)
-	    maxi = wi::floor_log2 (max);
-	}
-      if (mini == -2)
-	break;
-      r.set (build_int_cst (type, mini), build_int_cst (type, maxi));
-      return true;
-
     CASE_CFN_CLRSB:
       arg = gimple_call_arg (call, 0);
       prec = TYPE_PRECISION (TREE_TYPE (arg));
diff --git a/gcc/gimple-range-op.cc b/gcc/gimple-range-op.cc
index caba49309f9..801c2bb235e 100644
--- a/gcc/gimple-range-op.cc
+++ b/gcc/gimple-range-op.cc
@@ -489,6 +489,76 @@ cfn_clz::fold_range (irange &r, tree type, const irange &lh,
   return true;
 }
 
+// Implement range operator for CFN_BUILT_IN_CTZ
+class cfn_ctz : public range_operator
+{
+public:
+  cfn_ctz (bool internal) { m_gimple_call_internal_p = internal; }
+  using range_operator::fold_range;
+  virtual bool fold_range (irange &r, tree type, const irange &lh,
+			   const irange &, relation_kind) const;
+private:
+  bool m_gimple_call_internal_p;
+} op_cfn_ctz (false), op_cfn_ctz_internal (true);
+
+bool
+cfn_ctz::fold_range (irange &r, tree type, const irange &lh,
+		     const irange &, relation_kind) const
+{
+  if (lh.undefined_p ())
+    return false;
+  int prec = TYPE_PRECISION (lh.type ());
+  int mini = 0;
+  int maxi = prec - 1;
+  int zerov = 0;
+  scalar_int_mode mode = SCALAR_INT_TYPE_MODE (lh.type ());
+
+  if (m_gimple_call_internal_p)
+    {
+      if (optab_handler (ctz_optab, mode) != CODE_FOR_nothing
+	  && CTZ_DEFINED_VALUE_AT_ZERO (mode, zerov) == 2)
+	{
+	  // Handle only the two common values.
+	  if (zerov == -1)
+	    mini = -1;
+	  else if (zerov == prec)
+	    maxi = prec;
+	  else
+	    // Magic value to give up, unless we can prove arg is non-zero.
+	    mini = -2;
+	}
+    }
+  // If arg is non-zero, then use [0, prec - 1].
+  if (!range_includes_zero_p (&lh))
+    {
+      mini = 0;
+      maxi = prec - 1;
+    }
+  // If some high bits are known to be zero, we can decrease
+  // the maximum.
+  wide_int max = lh.upper_bound ();
+  if (max == 0)
+    {
+      // Argument is [0, 0].  If CTZ_DEFINED_VALUE_AT_ZERO
+      // is 2 with value -1 or prec, return [-1, -1] or [prec, prec].
+      // Otherwise ignore the range.
+      if (mini == -1)
+	maxi = -1;
+      else if (maxi == prec)
+	mini = prec;
+    }
+  // If value at zero is prec and 0 is in the range, we can't lower
+  // the upper bound.  We could create two separate ranges though,
+  // [0,floor_log2(max)][prec,prec] though.
+  else if (maxi != prec)
+    maxi = wi::floor_log2 (max);
+
+  if (mini == -2)
+    return false;
+  r.set (build_int_cst (type, mini), build_int_cst (type, maxi));
+  return true;
+}
+
 // Set up a gimple_range_op_handler for any built in function which can be
 // supported via range-ops.
 
@@ -553,6 +623,15 @@ gimple_range_op_handler::maybe_builtin_call ()
 	m_int = &op_cfn_clz;
       break;
 
+    CASE_CFN_CTZ:
+      m_op1 = gimple_call_arg (call, 0);
+      m_valid = true;
+      if (gimple_call_internal_p (call))
+	m_int = &op_cfn_ctz_internal;
+      else
+	m_int = &op_cfn_ctz;
+      break;
+
     default:
       break;
     }
-- 
2.37.3


  parent reply	other threads:[~2022-09-22 19:05 UTC|newest]

Thread overview: 18+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-09-22 18:49 [PATCH 00/17] Move builtin functions " Andrew MacLeod
2022-09-22 18:53 ` [PATCH 01/17] Replace another snippet with a call to, gimple_range_ssa_names Andrew MacLeod
2022-09-22 18:55 ` [PATCH 02/17] Adjust range_op_handler to store the handler directly Andrew MacLeod
2022-09-22 18:56 ` [PATCH 03/17] Create gimple_range_op_handler in a new source file Andrew MacLeod
2022-09-22 18:58 ` [PATCH 04/17] Fix calc_op1 for undefined op2_range Andrew MacLeod
2022-09-22 18:59 ` [PATCH 05/17] Add missing float fold_range prototype for floats Andrew MacLeod
2022-09-22 19:00 ` [PATCH 06/17] Always check the return value of fold_range Andrew MacLeod
2022-09-22 19:01 ` [PATCH 07/17] Add range-ops support for builtin functions Andrew MacLeod
2022-09-22 19:02 ` [PATCH 08/17] Convert CFN_BUILT_IN_SIGNBIT to range-ops Andrew MacLeod
2022-09-22 19:05 ` [PATCH 09/17] Convert CFN_BUILT_IN_TOUPPER and TOLOWER " Andrew MacLeod
2022-09-22 19:05 ` [PATCH 10/17] Convert CFN_BUILT_FFS and CFN_POPCOUNT " Andrew MacLeod
2022-09-22 19:05 ` [PATCH 11/17] Convert CFN_CLZ builtins " Andrew MacLeod
2022-09-22 19:05 ` Andrew MacLeod [this message]
2022-09-22 19:06 ` [PATCH 13/17] Convert CFN_BUILT_IN_CLRSB " Andrew MacLeod
2022-09-22 19:06 ` [PATCH 14/17] Convert CFN_BUILT_IN_UBSAN_CHECK_* " Andrew MacLeod
2022-09-22 19:08 ` [PATCH 15/17] Convert CFN_BUILT_IN_STRLEN " Andrew MacLeod
2022-09-22 19:10 ` [PATCH 16/17] Convert CFN_BUILT_IN_GOACC_DIM_* " Andrew MacLeod
2022-09-22 19:10 ` [PATCH 17/17] Convert CFN_BUILT_IN_PARITY " Andrew MacLeod

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=622e4a48-eae4-600f-db3c-c478f537caa7@redhat.com \
    --to=amacleod@redhat.com \
    --cc=aldyh@redhat.com \
    --cc=gcc-patches@gcc.gnu.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).